On August 28, JIDU Auto unveiled its all-new V2.0 software at an AI driving conference themed “Evolution: AI Reshaping Intelligent Driving” in Chengdu. The newly launched software will debut on the JIDU 07 model. Powered by Baidu Apollo’s Level 4 autonomous driving atomic capabilities and safety framework, JIDU Auto has achieved the milestone to offer pure vision technology for advanced intelligent driving, which is China’s exclusive and one of only two such offerings worldwide, maintaining the company’s leading position in the industry.
The V2.0 software is the most significant update for all JIDU Auto models, encompassing over a dozen key modules for up to 947 new features including end-to-end large-scale model smart driving, a large-scale model smart cockpit, and intelligent vehicle connectivity. Following the upgrade from PPA intelligent driving to ASD (Apollo Self Driving), JIDU Auto has taken a major step towards the era of end-to-end large-scale models on its pure vision technology for advanced intelligent driving, positioning the company as the first global smart EV brand to adopt Baidu Apollo ADFM—the Level 4 autonomous driving large-scale model.
“This marks a significant achievement as Level 4 autonomous driving technology is integrated into consumer cars for the first time. Thanks to the deep integration between Baidu’s ADFM and JIDU Auto, this is the world’s first large-scale autonomous driving model supporting fully unmanned driving applications,” said Joe Xia, CEO of JIDU Auto. “Baidu’s 10+ years of continuous R&D in AI and autonomous driving, coupled with investments exceeding RMB150 billion, has been instrumental in maintaining JIDU Auto’s leadership position.”
JIDU Auto’s fully-upgraded ASD: A giant leap in pure vision end-to-end technology
2024 is poised to be a breakthrough year for intelligent driving technology, as it increasingly becomes a key factor for consumers when choosing cars. End-to-end large models utilize two distinct data processing and transmission methods: one-stage and two-stage. Tesla’s one-stage approach prioritizes lossless transmission and direct information processing. However, this method presents challenges like unpredictable outputs, high debugging costs, and difficulties in guaranteeing a consistent minimum performance level. While the two-stage end-to-end model simplifies system design and implementation by processing data in phases, separating perception and decision-making. Each component performs its specific function, making test and maintenance more straightforward. By processing data in segments, it also allows for optimization at each stage, contributing to the overall system’s stability.
Considering safety, user experience, and other factors, JIDU Auto has opted for a two-stage end-to-end approach. While both JIDU Auto’s ASD and Tesla’s FSD are intelligent driving assistance systems based on pure vision technology, ASD is developed in China and has a deeper understanding of Chinese road conditions, which makes ASD a more suitable intelligent driving assistance system for the country’s specific traffic environment.
“Businesses will make their decisions between one-stage and two-stage models based on specific needs, recognizing the tangible benefits that end-to-end models offer. For us, the potential of intelligent driving technology is full of surprises,” said Wang Liang, Chair of Baidu Intelligent Driving Group Technical Committee. “Theoretically, using one-stage model to address the lower bound challenges is more difficult, requiring us to collect an exceptionally large amount of data with fast growth at an exponential rate compared to the two-stage model, as well as manage a large scale model and handle large amount of data.”
With the newly launched V2.0 software, JIDU Auto’s ASD has achieved a breakthrough by combining pure vision with an end-to-end large-scale model, which places ASD a generation ahead of the industry. Leveraging the Level 4 autonomous driving model Apollo ADFM, ASD strikes a balance between safety and generalizability, offering more safety to human drivers and encompasses the entire range of high-speed and complex urban scenarios.
Baidu is a global leader in AI large-scale models and autonomous driving technology, with over a decade of experience in AI and autonomous driving. Its technological capabilities are among the best in the world. Currently, Apollo’s Level 4 autonomous driving has surpassed 100 million kilometers in mileage and has established a high-performance training cluster with over 5.5 EFLOPS of computing power. Based on Joe Xia’s introduction, JIDU Auto will utilize a high-performance computing cluster exceeding 5.5 EFLOPS to enable efficient model training and rapid iteration. Meanwhile, through its “monthly evolution” approach, JIDU Auto will deliver increasingly advanced, intelligent, and safe intelligent driving experiences.
JIDU Auto’s pure vision technology for advanced intelligent driving surpasses mainstream LiDAR technology in perception capabilities. It achieves centimeter-level 3D precision, offering a perception range of up to 250 meters. Remarkably, it can even identify a mineral water bottle at a distance of 200 meters and maintains a recognition accuracy of moving obstacle speeds within 0.1 meters per second. JIDU Auto’s ASD introduces 106 new features and upgrades, including pilot assistance, valet parking, advanced lane keeping, and active safety, enhancing driving, safety, and parking capabilities.
Large-scale model smart cockpit evolves again: industry’s first lip movement recognition
At the press conference, JIDU Auto also unveiled the industry’s first native AI large-scale model cockpit. Built on an all-new native AI architecture, this cockpit redefines core functionalities by incorporating an innovative end-cloud AI agent system. Perception, comprehension, and execution are now model-driven, enabling both rapid local responses and leveraging the cloud’s powerful understanding and decision-making capabilities.
As early as October 2023, SIMO led the way by integrating Baidu’s ERNIE Bot, becoming the world’s first voice assistant powered by AI large-scale model. This enabled control of nearly all vehicle functions, delivering a more human-like and natural interaction. The newly upgraded SIMO 2.0 is even smarter and more human-like, revolutionizing the traditional command-based interaction of voice assistants. SIMO 2.0 boasts capabilities such as exploration, role-playing, and scene modes. It can assume the role of a family tutor, a car expert, or a travel guide, and it supports over 10,000 user-defined scene settings. Integrated with a multi-modal large-scale model, SIMO 2.0 achieved the industry’s first lip movement recognition function. By recognizing the user’s lip movement trajectory, speech recognition accuracy in noisy environments is increased to 90%. With the addition of features such as simultaneous recognition of multiple commands, multi-zone group chats, and voice commands inside and outside the car, it creates a truly immersive and futuristic intelligent cockpit experience.
As a premium intelligent automotive robot brand, JIDU Auto is always at the forefront of AI advancements, setting new standards for the future of AI mobility.
Explore AITechPark for the latest advancements in AI, IOT, Cybersecurity, AITech News, and insightful updates from industry experts!
